基于深度学习的儿童先天性巨结肠辅助诊断算法  

Deep Learning-based Algorithm for Computer-aided Diagnosis of Hirschsprung's Disease in Children

作　　者：李榕[1] 王庆煜 赵鑫申 贾子健 吴滢[2] LI Rong;WANG Qingyu;ZHAO Xinshen;JIA Zijian;WU Ying(School of Health Science and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Pathology Department,Shanghai Children′s Hospital,School of Medicine,Shanghai Jiao Tong University,Shanghai 200062,China)

机构地区：[1]上海理工大学健康科学与工程学院,上海200093 [2]上海交通大学医学院附属儿童医院病理科,上海200062

出　　处：《软件工程》2025年第4期26-31,共6页Software Engineering

基　　金：国家自然科学基金项目(82171696,82371711)。

摘　　要：在先天性巨结肠(Hirschsprung Disease,HD)的诊断工作中,神经节细胞的寻找耗时长且具有挑战性。为了更好地辅助病理医生诊断HD,提出一种基于改进YOLOv5s的结肠神经节细胞智能检测算法。该算法在YOLOv5s的Neck层PAN网络的3个C3模块后,分别添加了无参数的SimAM注意力机制,这一改进在不增加模型参数量的同时,有效地捕捉到更多重要的特征。实验结果显示,改进的算法能够快速且准确地识别神经节细胞,其精确度、召回率、mAP@0.5分别达到了87.4%、84.6%、91.5%。基于此算法,进一步开发了儿童先天性巨结肠辅助诊断软件,该软件能够自动检测结肠活检切片中的神经节细胞,为病理医生诊断HD提供了极大的便利。In the diagnosis of Hirschsprung′s Disease(HD),identifying ganglion cells is time-consuming and challenging.To assist pathologists in diagnosing HD,an intelligent detection algorithm for colonic ganglion cells based on improved YOLOv5s(You Only Look Once v5s)is proposed.This algorithm integrates parameter-free SimAM(Similarity-based Attention Mechanism)following the three C3 modules in the PAN network of the Neck layer.The integration enables the capture of more crucial features without increasing the model parameters.Experimental results demonstrate that the improved algorithm facilitates rapid and accurate identification of ganglion cells,achieving Precision,Recall,and mAP@0.5 of 87.4%,84.6%,and 91.5%,respectively.Based on this algorithm,a software for computer-aided diagnosis of Hirschsprung's Disease in children has been further developed,capable of automatically detecting ganglion cells in colon biopsy sections,thereby providing significant convenience for pathologists in diagnosing HD.

关 键 词：深度学习 YOLOv5 SimAM 先天性巨结肠 辅助诊断 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]